Community tools#

Projects built by the community on top of docTR. Built something useful? Open a pull request to add it to this page, with a short description and what makes it unique.

docTR Labeler#

github.com/text2knowledge/docTR-Labeler

An annotation tool to create and correct OCR datasets for training and fine-tuning docTR models.

  • Polygon editing: draw and edit polygons around text regions, with automatic shape correction

  • Auto-annotation: suggestions and polygon refinement powered by OnnxTR

  • Fast workflow: keyboard shortcuts for selection, zoom, undo/redo and saving

  • Flexible: launch it with the doctr-labeler command or use its Python API

  • Private: everything runs locally, no account needed

OnnxTR#

github.com/felixdittrich92/OnnxTR

An ONNX Runtime backend for docTR models: the same interface, without PyTorch. See ONNX & OnnxTR.

  • Minimal dependencies: no PyTorch or TensorFlow

  • Fast inference: CPU, GPU and accelerators such as OpenVINO

  • Quantization: 8-bit models for a smaller memory footprint

  • Familiar API: onnxtr.models.ocr_predictor, just like docTR

  • Production ready: Docker images and Hugging Face Hub integration

docling-OCR-OnnxTR#

github.com/felixdittrich92/docling-OCR-OnnxTR

A plugin that brings the OnnxTR OCR engine to the Docling document parsing framework.

  • Native integration: use it in Docling pipelines through OnnxtrOcrOptions

  • Drop-in replacement for the other OCR engines of Docling

  • Configurable: model choice, languages, confidence thresholds, batch sizes, orientation correction and 8-bit models